High dimensional predictions of suicide risk in 4.2 million US Veterans using ensemble transfer learning.

We present an ensemble transfer learning method to predict suicide from Veterans Affairs (VA) electronic medical records (EMR). A diverse set of base models was trained to predict a binary outcome constructed from reported suicide, suicide attempt, and overdose diagnoses with varying choices of study design and prediction methodology. Each model used twenty cross-sectional and 190 longitudinal variables observed in eight time intervals covering 7.5 years prior to the time of prediction. Ensembles of seven base models were created and fine-tuned with ten variables expected to change with study design and outcome definition in order to predict suicide and combined outcome in a prospective cohort. The ensemble models achieved c-statistics of 0.73 on 2-year suicide risk and 0.83 on the combined outcome when predicting on a prospective cohort of [Formula: see text] 4.2 M veterans. The ensembles rely on nonlinear base models trained using a matched retrospective nested case-control (Rcc) study cohort and show good calibration across a diversity of subgroups, including risk strata, age, sex, race, and level of healthcare utilization. In addition, a linear Rcc base model provided a rich set of biological predictors, including indicators of suicide, substance use disorder, mental health diagnoses and treatments, hypoxia and vascular damage, and demographics.

Deep sequential neural network models improve stratification of suicide attempt risk among US veterans.

OBJECTIVE: To apply deep neural networks (DNNs) to longitudinal EHR data in order to predict suicide attempt risk among veterans. Local explainability techniques were used to provide explanations for each prediction with the goal of ultimately improving outreach and intervention efforts. MATERIALS AND METHODS: The DNNs fused demographic information with diagnostic, prescription, and procedure codes. Models were trained and tested on EHR data of approximately 500 000 US veterans: all veterans with recorded suicide attempts from April 1, 2005, through January 1, 2016, each paired with 5 veterans of the same age who did not attempt suicide. Shapley Additive Explanation (SHAP) values were calculated to provide explanations of DNN predictions. RESULTS: The DNNs outperformed logistic and linear regression models in predicting suicide attempts. After adjusting for the sampling technique, the convolutional neural network (CNN) model achieved a positive predictive value (PPV) of 0.54 for suicide attempts within 12 months by veterans in the top 0.1% risk tier. Explainability methods identified meaningful subgroups of high-risk veterans as well as key determinants of suicide attempt risk at both the group and individual level. DISCUSSION AND CONCLUSION: The deep learning methods employed in the present study have the potential to significantly enhance existing suicide risk models for veterans. These methods can also provide important clues to explore the relative value of long-term and short-term intervention strategies. Furthermore, the explainability methods utilized here could also be used to communicate to clinicians the key features which increase specific veterans' risk for attempting suicide.

Chief complaint classification with recurrent neural networks.

Syndromic surveillance detects and monitors individual and population health indicators through sources such as emergency department records. Automated classification of these records can improve outbreak detection speed and diagnosis accuracy. Current syndromic systems rely on hand-coded keyword-based methods to parse written fields and may benefit from the use of modern supervised-learning classifier models. In this paper, we implement two recurrent neural network models based on long short-term memory (LSTM) and gated recurrent unit (GRU) cells and compare them to two traditional bag-of-words classifiers: multinomial naïve Bayes (MNB) and a support vector machine (SVM). The MNB classifier is one of only two machine learning algorithms currently being used for syndromic surveillance. All four models are trained to predict diagnostic code groups as defined by Clinical Classification Software, first to predict from discharge diagnosis, and then from chief complaint fields. The classifiers are trained on 3.6 million de-identified emergency department records from a single United States jurisdiction. We compare performance of these models primarily using the F1 score, and we measure absolute model performance to determine which conditions are the most amenable to surveillance based on chief complaint alone. Using discharge diagnoses, the LSTM classifier performs best, though all models exhibit an F1 score above 96.00. Using chief complaints, the GRU performs best (F1 = 47.38), and MNB with bigrams performs worst (F1 = 39.40). We also note that certain syndrome types are easier to detect than others. For example, chief complaints using the GRU model predicts alcohol-related disorders well (F1 = 78.91) but predicts influenza poorly (F1 = 14.80). In all instances, the RNN models outperformed the bag-of-words classifiers suggesting deep learning models could substantially improve the automatic classification of unstructured text for syndromic surveillance.

A spatial model of the efficiency of T cell search in the influenza-infected lung.

Emerging strains of influenza, such as avian H5N1 and 2009 pandemic H1N1, are more virulent than seasonal H1N1 influenza, yet the underlying mechanisms for these differences are not well understood. Subtle differences in how a given strain interacts with the immune system are likely a key factor in determining virulence. One aspect of the interaction is the ability of T cells to locate the foci of the infection in time to prevent uncontrolled expansion. Here, we develop an agent based spatial model to focus on T cell migration from lymph nodes through the vascular system to sites of infection. We use our model to investigate whether different strains of influenza modulate this process. We calibrate the model using viral and chemokine secretion rates we measure in vitro together with values taken from literature. The spatial nature of the model reveals unique challenges for T cell recruitment that are not apparent in standard differential equation models. In this model comparing three influenza viruses, plaque expansion is governed primarily by the replication rate of the virus strain, and the efficiency of the T cell search-and-kill is limited by the density of infected epithelial cells in each plaque. Thus for each virus there is a different threshold of T cell search time above which recruited T cells are unable to control further expansion. Future models could use this relationship to more accurately predict control of the infection.

Higher level of replication efficiency of 2009 (H1N1) pandemic influenza virus than those of seasonal and avian strains: kinetics from epithelial cell culture and computational modeling.

The pathogenicity and transmission of influenza A viruses are likely determined in part by replication efficiency in human cells, which is the net effect of complex virus-host interactions. H5N1 avian, H1N1 seasonal, and H1N1 2009 pandemic influenza virus strains were compared by infecting human differentiated bronchial epithelial cells in air-liquid interface cultures at relatively low virus particle/cell ratios. Differential equation and computational models were used to characterize the in vitro kinetic behaviors of the three strains. The models were calibrated by fitting experimental data in order to estimate difficult-to-measure parameters. Both models found marked differences in the relative values of p, the virion production rate per cell, and R(0), an index of the spread of infection through the monolayer, with the values for the strains in the following rank order (from greatest to least): pandemic strain, followed by seasonal strain, followed by avian strain, as expected. In the differential equation model, which treats virus and cell populations as well mixed, R(0) and p varied proportionately for all 3 strains, consistent with a primary role for productivity. In the spatially explicit computational model, R(0) and p also varied proportionately except that R(0) derived for the pandemic strain was reduced, consistent with constrained viral spread imposed by multiple host defenses, including mucus and paracrine antiviral effects. This synergistic experimental-computational strategy provides relevant parameters for identifying and phenotyping potential pandemic strains.